It is tempting to think that the most trustworthy source is the one that is never wrong. No such source exists. Reporting happens under time pressure, with incomplete information, about events that are still unfolding. Mistakes are inevitable. The real question is what happens next.

Silent edits destroy trust

When an error is quietly fixed, with the headline changed and a number replaced without a note, readers who saw the original are left believing the mistake. Worse, anyone who noticed the change learns that the publication rewrites its record without telling anyone. The next time it is right, that reader will have less reason to believe it.

Visible corrections build it

A correction note, stating what was wrong, what is right, and when it was changed, does the opposite. It tells readers that the publication takes accuracy seriously enough to admit a failure in public. Over time, a visible record of corrections is one of the strongest signals of reliability a newsroom can offer.

An international standard

The International Fact-Checking Network, based at the Poynter Institute, asks its signatories to commit to a Code of Principles. Alongside nonpartisanship and fairness, transparency of sources, transparency of funding and organisation, and transparency of methodology, one of its core commitments is an open and honest corrections policy. It is not a courtesy; it is part of the definition of the job.

What a good correction looks like

  • It is visible, placed where readers of the original will see it.
  • It is specific about what was wrong and what is right.
  • It is timely, published as soon as the error is confirmed.
  • It is proportionate: a wrong headline needs a correction as prominent as the headline.

How this applies to Honest Lens

Honest Lens applies the same principle to its own engine. The Self-Audit page is built to publish where the system disagrees with human fact-checkers and why, instead of presenting only its successes. Readers can also flag an evaluation as too strict or missing an angle; that feedback is stored for review rather than discarded. An automated fact-checker that never admits error would be the least trustworthy kind.